Heuristic
Add planner-builder-evaluator roles only after the simple loop hits a ceiling
Three agents are not automatically one agent with better judgment.
When it fits
- A single-agent workflow is failing on decomposition or quality and you are considering a multi-role architecture.
When to avoid it
- Extra agents add latency, cost and new coordination failures; simplify again when a stronger model makes a role redundant.
Why it matters
Start with the simplest capable worker and a clear test. Add a planner when decomposition itself is a recurring failure, and add an evaluator when independent quality feedback improves results. Measure the new loop against the simpler baseline. Keep role boundaries narrow enough that handoffs carry artifacts and decisions rather than duplicated conversation.
An example
A full-stack build may justify planner, builder and evaluator roles; a five-line configuration change probably does not.
Check your result
Each added agent role exists because it fixes a measured failure mode, not because multi-agent architecture sounds advanced.
Keep this limit in mind
- Extra agents add latency, cost and new coordination failures; simplify again when a stronger model makes a role redundant.
Evidence and sources
Anthropic recommends increasing harness complexity only when simpler approaches fail, and used component removal to identify which harness pieces were load-bearing.
Ablation can miss interactions between components; test representative end-to-end tasks.
Harness design for long-running application development · Simplification and methodical component removal